Sarcasm Detection: A Systematic Review of Methods and Approaches

Abstract

Social media is a common source of communication for various formal and informal contextual use cases. The conversation in both structured and unstructured forms can be broadly classified as positive/negative. In addition to “sarcasm,” the research about unstructured language has become very interesting due to the fact that very few researchers have offered solutions to problems associated with it. By using deep learning models, some hybrid approaches are used to identify sarcasm sentences. The identification is further refined to mark the content as sarcasm, irony, humour and offensive. This article analyzes and summarizes various works on irony/sarcasm detection in terms of features, approach, architecture and performance. This study analyzed that, the hybrid models superseded the performance of the traditional machine learning approaches for classifying the sarcasm/irony content. Finally, this study has briefed the identified challenges and research directions for building better models for classifying sarcasm/irony content.